One Command Back In

tmux-scoped Amplifier sessions, 1833 lines

Getting into a project is the same chore, every time — one command erases it

Open VS Code, make a terminal, cd into the repo, type amplifier. amplifier-workspace collapses that ceremony into amp <project>.

And the whole thing is smaller than you'd guess.

# the old way, every project open VS Code open a terminal cd ~/dev/ANext/amplifier-core amplifier # the new way amp core
Status: Shipped

The entire system is one command's worth of glue

~1833 lines across the tracked source and docs, landed in a single commit by Sam Schillace on 2026-02-25. Not vaporware — a small, finished tool.

Small fix, but the pain it targets is real.

1833
total lines (wc -l)
1
commit (1d5c509)
13
files changed
341
lines: largest unit, amp.sh

Repeat that ritual across a dozen workstreams — then lose it to a crash

The design targets Sam juggling 5–10 things in parallel. Every switch pays the setup tax, and a flaky VS Code + WSL setup can wipe the state you were holding.

So the tool replaces the whole sequence with one word.

5–10
parallel workstreams (DESIGN.md)
58
projects, ~16 active (documented background)

One partial name resolves to the right directory and a dedicated session

amp core fuzzy-resolves to amplifier-core, then creates or attaches a tmux session named amp-<project> and drops you straight into amplifier.

And it doesn't just open — it remembers.

If a prior session is on disk, amp continues where you left off

amp-session.sh reads ~/.amplifier/projects/<slug>/sessions. If any exist it runs amplifier continue; if not, it starts fresh — so switching back restores context.

Which means there's nothing to track and nothing to lose.

# amp-session.sh SESSIONS_DIR=~/.amplifier/projects/$SLUG/sessions # found existing sessions? amplifier continue # resume # none found? amplifier # fresh

No list to maintain, nothing to lose — the system tracks and survives on its own

Active projects are detected automatically from git commits in the last 14 days. tmux-resurrect + continuum auto-save every 15 min and auto-restore, so sessions survive WSL crashes.

And the same command reaches beyond your laptop.

The same command reaches the whole fleet — local and remote are one move

spark <project> just SSHes to a fleet node (default spark-1) and runs the identical amp there. --on <node> targets a specific machine.

And every machine gets there the same way.

# run amp on the default fleet node spark core # becomes, over SSH: ssh -t spark-1 "amp core" # target a specific machine spark --on spark-2 core

One idempotent setup.sh makes every machine speak the same amp

A single setup script wires the whole workspace in six numbered steps — safe to run repeatedly. Install once per machine, and the frictionless entry is identical everywhere.

One command in; the same command, everywhere.

Sources

Research Methodology

Data as of: 2026-02-25 (single commit 1d5c509)  ·  Feature status: Shipped

Repository: ramparte/amplifier-workspace (singular) — not the unrelated amplifier-workspaces web app

Commands run (independently re-derived in a fresh session):

Contributor: Sam Schillace — sole author of the only commit; co-authored-by Amplifier (generated with microsoft/amplifier).

Gaps / caveats: The '58 projects (~16 active)' figure is DESIGN.md narrative (Sam's self-description), not a live filesystem count. wc -l reports 1833 vs git's 1835 insertions — a trailing-newline artifact; both correct for their commands. amp --new does mkdir + git init + launch only (no repo cloning, no AGENTS.md/settings.yaml generation).

More Amplifier Stories